Super-Yi Dual Opteron rack-type Server evaluation

Server

The super-Yi customization Server includes a series and I series, the A series is a server based on the AMD series processor, and the I series is a server based on the Intel family of processors. Our evaluation is the use of dual AMD Opertron processor 220A-2806T Server, it is located in the enterprise-class core overload dedicated server, small and medium-sized storage multi-function server, Internet E-commerce Network operation of the core server, It's the same thing that we did before we evaluated 220a-2806t.

The customizer 220a-2806t server looks exactly the same as the 220x-2806vs server, and the appearance of the Supermicro superserver 6024 server is very similar. 2806T is 2U rack-type structure, the overall size is 89mm (high) x 426mm (wide) x 650mm (deep), its depth is shorter than most of the server.

2806T front panel design has power switch, reset switch, Power LED, hard drive indicator, two NIC status indicator, power failure led, overheat/fan fault led. The front panel is also designed with 6 hard drive bays, which are configured with 3 Seagate Barrucade 7200.7 120GB SATA interface hard drives and a RAID 5 mode disk array with ARC1120 SATA RAID cards. The optical drive device is a normal CD-ROM drive. There is no USB interface and no positioning indicator on the front panel, which will be slightly inconvenient to use.

Super-yi 2806VS server Back IO panel provides 1 PS/2 mouse interface, 1 PS/2 keyboard interface, a serial port, a VGA interface, a Gigabit network interface (Intel 82551QM 10/100 LOM), 22 Gigabit Nic Interface (RJ45, Broadcom bcm5704c Gigabit NIC Controller) and 2 USB 1.1 ports. This server uses the single power supply design, does not provide the redundant power expansion ability, the power supply model is ablecom sp552-2c, the maximum output power can reach 550W, its basic index is as follows table:

35a

+3.3v	+5v	+12v	-12v	-5v	+5VSB
30a	42a	1.0a	0.8a	2a
+3.3v and +5v merge maximum current 50a

This server employs the Tyan S2882 motherboard, which uses the AMD-8131 hypertransport pci-x tunnel+amd-8111 I/O hub chipset, with the largest support for 2 AMD Opteron processor (UPGA 940-pin). Because the Opteron processor consolidates the memory controller, the memory DIMMs on the motherboard are divided into two groups, each of which is controlled by a single processor memory controller. 4.

Two Opteron 246 processors (2GHZ,1MB L2 cache) are configured on the test server, which consolidates the 128bit DDR memory controller, which supports a dual-channel DDR memory configuration with a maximum of 16GB REG ECC ddr400/333/266 memory. With a single 1GB DDR266 memory (Samsung PC2100 CL2.5) configured for each processor, the duo cannot perform the dual-channel memory controller, and if it is actually used, we recommend that the user use two 512MB memory instead of a single 1GB.

The chassis is built with 4 cooling fans, running at speeds above 6000RPM. The middle two chassis cooling fan is primarily used for heat dissipation of the processor and memory because the shroud is not designed so that a fan is installed on each processor's heatsink to ensure that the processor is adequately cooled.

5 PCI expansion slots are available on the motherboard, including 4 pci-x slots and a 32bit 33MHz PCI slot. The two pci-x on the left side of the picture above are controlled by 1 Pci-x Bridge, the right two pci-x have another 1 Pci-x bridge control, the rightmost slot can support the ZCR expansion card. The Silicon Image Sil 3114 SATA Controller supports 4 SATA150 devices while SATA raid (RAID 0, 1, 0+1) is supported. This server provides 6 3.5-inch hard drive bays, so it's clear that using onboard RAID controllers is not enough, so the super-Pci-x has a ARC1120 SATA RAID card with the Intel 80331 I/O processor, 128MB cache, You can support more modes, such as RAID 0/1/3/5/6/jbod, and can connect up to 8 SATA devices.

Server installation, maintenance and management Customizer 220x-2806vs Server to provide driver CD-ROM, power cord and other accessories, we do not see any documentation, perhaps the manufacturer does not include in the delivery test products, we will further verify. From the appearance of the server, we judged that the installation of their rails is very traditional, through screws to the internal rail fixed on both sides of the chassis, and then the external rail fixed on the standard rack (depth of not less than 650mm) to complete the installation.

Random Disc

The tool CD of the super-Yi server is based on the server board that it uses, choose a different brand of motherboard, come with the tool CD that comes with the different brand motherboard. The drive disc for the Tyan S2882 motherboard does not have a boot feature and is primarily a driver for the Tyan series motherboard.

Tyan SYSTEM MONITOR v2.2.2

Tyan SYSTEM Monitor is also C/s mode, install the agent program on the server being monitored, install console program on the console, can monitor the running state of the server remotely.

The Tyan System Monitor program monitors the speed of processor fans and chassis fans, CPU core voltages, System 3.3v/5v/12v voltages, DDR vtt/2.5v voltages, and alerts if more than set thresholds are exceeded.

The Tyan SYSTEM Monitor program can send alerts to a specified address via a pre-set email server and generate OEC logs.

Graphical Monitoring interface

Tyan SYSTEM Monitor program's biggest feature is to provide intuitive graphical monitoring tools tyanscope, this tool can monitor the processor, processing VRM, memory, each fan and so on, as long as the mouse click the appropriate location, you can visually view the situation of the monitored equipment.

The motherboard used by the server can connect the Tyan server management to support the IPMI 1.5 specification, which can be used in conjunction with other management tools that support this specification.

Overall, the super-Yi server only provides a relatively basic server monitoring function, this gap with other domestic manufacturers there is a big gap.

Test platform and test methods 2005, is the SME information active year, but also the most Active server procurement year. Vendors have introduced server products for SMB applications, and it is not easy for small and medium enterprises to choose their own products in a wide range of brands and models. Therefore, this time we take the form of "propositional composition" to the manufacturer to solicit products: located in the SMB application of the rack-type server, the price between 230,000.

All of the test manufacturers for our evaluation of product requirements for the careful selection and configuration, so we tested each of the products are in fact by various manufacturers specifically for small and medium-sized applications customized server. Compared with the standard configuration, these products have better balance, availability and cost performance.

Our evaluation is not limited to performance testing, mainly includes the following 6 aspects:

I. Evaluation of MAIN SPECIFICATIONS

This section is mainly for the test server's physical characteristics, hardware configuration, and so on, such as Server chassis architecture, specification size, processor model/quantity/frequency, chipset, memory model and capacity, disk controller, hard disk model, disk array, network card model and interface type, thermal device, power supply, expansion slot, Io interface, number of bays, and so on.

Second, the Server technology verification test

Processor technology: SMP technology, HT Technology, EMT64, Dual Core

Memory: DDR, Ddrii, dual channel, memory mirroring, memory hot standby

Disk interface technology: SATA, IDE, SCSI

Disk array technology: RAID 0, RAID 1, RAID 10, RAID 5 Hot Swap Technology

Mainstream operating system management capabilities, Windows, Linux Local management, and remote management

System Configuration and Management Service configuration

Hardware monitoring and Management: CPU, disk, network log management, alarm

SEPCCPU2000 can evaluate the integer performance and floating-point performance of the processors in the server system, and we are testing the specint_base2000 and specfp_base2000 two projects. Sciencemark v2.0, a software that evaluates the capabilities of the processor's scientific computing, provides the ability to evaluate the memory bandwidth of the system being tested, which can support multithreading very well. Iometer is an I/O subsystem evaluation tool that can be used for a single system or cluster system, and we only use its disk testing capabilities. Webbench and NetBench are two commonly used software in server evaluation, which focuses on the performance evaluation of the server CPU subsystem, while the latter focuses on the performance evaluation of the IO subsystem. We installed the Microsoft Windows 2003 Standard Edition (Chinese, 32bit) on the server above, installed the driver of each hardware correctly, optimized it, closed the unwanted service and ensured the server was in the best state. In previous server evaluations, we generally used spec CPU2000 to evaluate the floating-point performance and integer performance of the processor. However, since we lack the necessary compilers to support the AMD platform to run spec cpu2000, we have to give up the software in this part, using the more Easy-to-use sisoft Sandra Pro.

In the arithmetic operation Benchmark and the multimedia benchmark test, the performance of the platform based on the dual Xeon is higher than the dual Opteron platform. Because SiSoft Sandra 2005 is a well supported test software for multithreading, it is perfectly normal to lead a dual-Xeon platform that supports hyper-Threading technology in processor testing.

	220X	220A
Memory bandwidth	4031.20 MB/s	1928 MB/s
L1 Cache Latency
Bytes Stride	3 Cycles/1.00ns	3 Cycles/1.00ns
L2 Cache Latency
4 Bytes Stride	6 cycles/2.00 NS	3 cycles/1.00 NS
Bytes Stride	One cycles/3.67 NS	5 cycles/1.67 NS
Stride Bytes	cycles/9.00 NS	cycles/5.67 NS
256 Bytes Stride	Num cycles/8.67 NS	cycles/4.00 NS
Stride Bytes	cycles/8.33 NS	cycles/4.33 NS
Memory Latency
4 Bytes Stride	6 cycles/2.00 NS	5 cycles/1.67 NS
Bytes Stride	cycles/5.00 NS	cycles/6.33 NS
Stride Bytes	cycles/16.33 NS	cycles/24.33 NS
256 Bytes Stride	379 cycles/126.33 NS	147 cycles/49.00 NS
Stride Bytes	398 cycles/132.67 NS	158 cycles/52.67 NS
Algorithm Bandwidth
Compiler	2023.26 MB/s	13973.56 MB/s
REP Movsd	2066.17 MB/s	14561.53 MB/s
ALU Reg Copy	1807.30 MB/s	8247.49 MB/s
MMX Reg Copy	1934.88 MB/s	20032.86 MB/s
MMX Reg 3dNow	-	-
MMX Reg SSE	3166.06 MB/s	-
SSE palign	3438.64 MB/s	15964.52 MB/s
SSE Palign SSE	3360.15 MB/s	-
SSE2 palign	3427.43 MB/s	14139.8 MB/s
SSE2 palign SSE	3357.33 MB/s	-
MMX Block 4KB	3574.19 MB/s	-
MMX Block 16kb	4031.20 MB/s	-
SSE Block 4KB	3522.50 MB/s	-
SSE Block 16kb	4020.42 MB/s	-

Sciencemark v2.0 is a software used to test the performance of the system, especially the processor in a scientific computing application, Membenchmark is a functional module designed for processor caching and system memory that can test system memory bandwidth, L1 cache latency, L2 Cache latency and system memory latency, in addition to testing the performance differences of different instruction sets.

The Super 2806VS server uses the E7520 chipset, its memory controller supports dual channel Ddrii memory, Ddrii 400 (3-3-3-8) memory is used in the test configuration, and the memory controller of the super-Yi 2806T Server is integrated into the processor. The test results show that the L2 cache latency and memory latency are the obvious advantages of the AMD platform. AMD can run algorithm bandwidth test projects, but the performance of the test results are significantly ahead.

It can be seen that different testing software for different platforms to provide different degrees of optimization, so we only from the actual application test to see in the end that the overall performance of the platform is more excellent.

Disk Performance test: The IOMeter 2004.07.30IOMeter is a very powerful IO test software that provides the ability to simulate network applications in addition to the IO (disk) performance of the computer running the test native. In this test, we just let it run the test server's disk performance on the local computer. In order to fully test the IO performance of the server being tested, we selected different test scripts separately. Max_throughput: File size is 64kb,100% read operation, the random rate is 0%, used to detect the maximum throughput of disk system Max_io: File size for 512b,100% read operation, random rate of 0%, Maximum IO capability for detecting disk systems Fielserver: File size ranging from 0.5KB to 64KB, 80% read operations, random rate of 100%, for analog file server performance webserver: File size from 0.5KB to 512KB, 100% read operations , with a random rate of 100%, used to simulate the performance of a Web server

The 220X-2806VS server uses a separate lsi-320-1 LP pci-x raid control card and 3 Seagate Cheetach 10k.7 146GB hard drives to form a RAID 5 mode disk array, when the queue depth is 128, IO processing capability can reach IOs 40732 per second. The super-220A-2806T server employs the ARC-1120 SATA pci-x RAID Controller and 3 Seagate Barracuda72000.7 120GB hard drives, comprising a RAID 5 mode array with a maximum IO processing capacity of around 22000, Under different queue depth conditions, the performance is very smooth.

The 220X-2806VS server disk subsystem has a maximum throughput of 140mb/s (8/16/32), although the throughput is very low under other queue depth conditions. The throughput of the 220A-2806T server disk subsystem is always maintained between 60-80mb/s, and the throughput does not fluctuate wildly at different queue depths.

Network connectivity Performance: NetBench 7.03NetBench is for file server performance testing software, affecting NetBench performance is mainly the server's disk subsystem, server disk controller, hard disk type, build disk array mode will have a significant impact on the test results. We set up a file server on the tested server, NetBench through 32 clients in the network lab to simulate the file transfer request from the PC to the file server, the file server sends the file data stored on disk to the corresponding client. With the built-in NIC.TST test script, we can test the throughput and response time of the server Gigabit NIC: The 220X-2806VS server board contains the Broadcom BCM5704 dual-channel Gigabit NIC Controller, which can be expected to provide almost no difference in performance of the two gigabit network. From our test results, both throughput and response time, the difference between the two is negligible. However, we have carefully compared the results of the two, select one of the better results of a network of the following network-related testing projects. Webbench is testing the performance of the server as a Web server, we have installed the IIS6.0 component on the tested server to provide the Web services required for testing. In the test we opened the network Lab 32 PCs (Celeron 1.7/256mb/40gb/100ethernet) as the client, respectively, using the Webbench 5.0 built-in dynamic CGI and static page script to test the server. Static testing (STATIC.TST) is a Web page (Wbtree) that is read by the client in advance of a server Web server, and this test focuses on server disk system and network connectivity performance. We used all 32 PCs in the lab to send requests to the tested server, and the test curve of the 220A platform was significantly better than 220X. Of course 32 Client access, 220X server processing capacity reached the 10005 requests/s, and 220A reached the 11350 requests/s. In multithreaded static script testing (STATIC_MT.TST), each physical client emits two threads, and when 32 clients, 220X server processing power reaches the 13350 requests/s,220a server's processing capacity to 13289 requests/s, Very close. Dynamic testing focuses on performance testing of the server's CPU subsystem, which provides enough load for the Web server. We compile the CGI source file simcigi.c in a C language into Simcgi.exe and use it as a CGI script in the dynamic test. During the testing process, each set up weThe Bbench client software's PC will continue to send CGI requests to the server within 300 seconds, while the console records and summarizes the data that the server responds to the CGI request.   When the Webbench test is started, the client connects to the server sequentially and sends a CGI request in each of the four sets. When the test is finished, the console collects the data and draws the change curve of the number of server CGI responses. The performance of the CGI test is mainly determined by the quality of the server processor subsystem. The processor subsystem includes CPU, memory, and memory controller, CPU frequency, cache, memory size, and memory bandwidth, which can affect the performance. From the test results curve, when there are 32 clients connected to the tested server at the same time, 220X Server processing capacity of 3867 requests/s,requests/s, the performance gap between the two is only about 7%, far from the front sisoft Sandra 2005 such a synthetic software test shows that there is such a big gap. NetBench is the performance testing software for file servers, the main impact of NetBench performance is the server's disk subsystem, the server disk controller, the type of hard disk, the build disk array mode will have a significant impact on the test results. We set up a file server on the tested server, NetBench through 32 clients in the network lab to simulate the file transfer request from the PC to the file server, the file server sends the file data stored on disk to the corresponding client. As with the Webbench test, the client will connect to the server and send file transfer requests in each of the four sets after the NetBench test begins. At the end of the test, the console collects the data and draws the change curve of the data transfer that the server can achieve.   Before we tested the 220X-2806VS server, we found that the server was very depressed in the NetBench test, and it was quite different from the previous one, which was the standard disk for the double Xeon platform. The mix test (DM.TST) has a maximum throughput of only 123 Mbps (32 clients), Enterprise Disk Mix Test (ENT_DM. TST) tested maximum throughput of only 207Mbps (12 clients). The performance of the 220A-2806T server, like the single test, is still very smooth, and the standard disk Mix test (DM.TST) has the highest throughput of 167 Mbps (32 clients), Enterprise Disk Mix tests (ENT_DM. TST) tested the highest throughput reached 439 Mbps. The delivery test product upgrades the processor from 244 to 246, the number increases to 2, and the memory is increased from standard 512MB to 2GB, the hard drive reform increased to 3 blocks, made up of RAID 5, and the overall price reached 29799 yuan. Ultra-yi Sales of AMD Opteron 244 processor quotes for 1780 Yuan, AMD Opteron 246 processor quote for 2360 Yuan. Because the AMD 8000 chipset is a historically "long" product, the main drawback of the motherboard products based on it is inadequate support for new technologies and specifications. For example, it still uses DDR memory, cannot support the PCI-E expansion card, and the USB port is still version 1.1. Fortunately, these problems in the server will not have a significant impact on the normal use. Ultra-Yi does not have self-developed management software, so the customizer 220a-2806t server Tyan SYSTEM monitor 2.2.2 Software, it can provide some basic remote monitoring function.

Single processor, memory and disk performance test shows that 220A server performance is flat, with the double Xeon Flat 220X has a big gap. However, the gap between the two is greatly reduced in the test of practical applications such as Webbench and NetBench, and even part of the test results in a 220A lead.

（